GPT 4 vs GPT 5: Here’s what’s improved and what’s not so great

On Thursday, OpenAI unveiled its much-anticipated GPT-5, marking a significant upgrade in its series of language models. This new iteration is now available to all users, replacing older versions such as GPT-4o and others that previously powered ChatGPT. According to OpenAI, GPT-5 represents a major leap in capabilities compared to its predecessors. The model is touted to excel in various domains, including coding, mathematics, writing, health inquiries, and visual interpretation. OpenAI's CEO, Sam Altman, referred to GPT-5 as a 'PhD-level expert' across these fields, highlighting its advanced performance. One of the standout features of GPT-5 is its enhanced speed and accuracy. The company claims that the new model is 45% less likely to generate factual inaccuracies compared to GPT-4o and 80% less so than its earlier model, o3. However, users on social media have expressed mixed feelings, with some reporting that responses seem slower and less accurate than those generated by GPT-4o. In terms of coding capabilities, GPT-5 is described as OpenAI's most proficient model yet. It can generate complex front-end applications and debug extensive codebases from simple natural language prompts, significantly lowering the barrier for users without coding experience. When it comes to writing assistance, OpenAI claims GPT-5 is the 'most capable writing collaborator' to date. While some users have noted improvements in narrative consistency and flow, others have criticized the model for being less creative than its predecessor, indicating issues with character consistency and emotional depth in its responses. OpenAI has also made strides in health-related queries. The new model is designed to engage users proactively, offering insights and raising potential health concerns, thereby acting as a more interactive thought partner. This adaptability is said to improve the model's contextual understanding and safety in providing advice. In a significant change, OpenAI has removed the model selection feature from ChatGPT, meaning users can no longer choose between different versions. Instead, GPT-5 operates with an efficiency model and a reasoning model, with a real-time router deciding which to use based on the conversation's complexity and intent. Usage limits for GPT-5 remain similar to previous models, allowing free-tier users to send 10 messages every five hours, while Plus and Pro users have increased limits. In response to user feedback about the removal of older models, OpenAI has temporarily doubled the message limit for Plus users to 160 messages every three hours. Despite improvements, some long-time users have reported that GPT-5 lacks the engaging personality they were accustomed to, describing it as more sterile and formal. Altman has acknowledged these concerns and mentioned that OpenAI plans to enhance the model's warmth in future updates.